Thread by @steveparks

Today I’m at #BBCtechconf in Broadcasting House, a gathering of BBC staff working on leading edge tech and external experts. I’ll cover some of the discussions in this thread...

The first session is Who’s AI Is It Anyway - but in the session was setup with an interesting provocation: could AI replace government? The panel includes @IndraJoshi10 from NHSX. She talks about AI being more about supporting rather than repacking...

She describes a consortium in the NHS that is exploring having AI supporting the interpretation of mammograms.
Another panellist says there is serious potential with these technologies - citing Ocado reducing food waste by 5% by using AI

But says we must take a pragmatic approach to AI so it is for the betterment of society. Consider bias, consider adverse effects. But what worries her is the countries advancing on AI but with less attention to ethics - because the tech is borderless

Tina, the presenter, moves the conversation to the fuel that drives AI - Data. She shows a clip from Killing Eve in which someone reveals how much they have been able to know about a person from their data

Panel member @SandraWachter5 talks about the data we already provide through social media - and that there are studies already into how much targeting through that data is ethical. Eg if to target someone with a gambling problem to gamble

But then there can be the positive uses - your Fitbit buzzing to let you know to move around more. So there is lots to explore and think about with the very important issues around ethics

Apologies - that ^ was @sanakb. No name badges!
Tina asks for examples of good practice. Sana says that government has created an online harms white paper, and a lot of tech companies have ethical codes. Says people are thinking about this, especially post Cambridge Analytica

She says government is giving funding to open research institutes. Also there is the Alan Turing institute where a main priority is safe and ethical AI research. “Gone are the days when the big tech companies are doing some schmancy thing and we all just accept.”

She says we’re now more award and able to provide more informed consent (not sure I agree with this!!!)

Indra says that in health there is the duty of confidentiality, on top of that there’s GDPR so there is a lot of regulation.

Indra says that the key is to show people the positive outcomes - if the NHS shows that by doing this this and this with their data the benefit could be x. It’s important to understand the bigger picture

Sana is asked if it’s too late to regulate AI such as facial recognition. She says it’s not too late, but it is already with us every day - eg using our face to unlock our phones or banking apps instead of writing down our passwords

But then what if facial recognition is used by a foreign government for ethnic cleansing? There’s clearly a line that’s crossed. So let’s have a conversation on that line, but let’s focus our discussions around the middle rather than the extremes

There’s now a short film about the emergence of bias in AI algorithms - showing black women identified as male by the AI. And Oprah Winfrey being identified as Michelle Obama

They show many examples in multiple AI tools, including IBM Watson.
They refer to notflawless.ai to see examples

Sandra says it starts with the data. And getting diversity into the design of the data gathering and processing is key. Diversity of thought, of literature and more

Indra emphasises that we all have biases, and that gets through into the tech. What we have to do is call it out when we see it. It’s up to all of us.
Sana says the problem of diversity in tech is getting worse - we really need to fix this and we need to fix it together.

She says there are programmes of funding now for STEM conversion courses to support women and other under-represented demographics moving into technology study at later points. And AI is attracting more women because it’s about solving big world problems

Now discussing how much regulation is needed. Indra says there are two types of regulation - hard laws, and softer ethical considerations. The latter falls on all of us
Sandra says some can be codes of conduct, some can be regulation

Tina asks how important it is to be able to explain exactly how the algorithms work and what they are doing. Indra says in health there is a drug they use to put people to sleep...

...they’re not exactly sure of the pharmakinetixs of that drug - but there have been lots of controlled studies to understand the doses, the effects and so on to understand it’s safe usage. She says AI is similar. We don’t need to see the exact code all the time...

...but we do need to understand the inputs, actions and outputs at a high level.
There are many wide-ranging uses - better distribution of blood supplies, managing the flow of traffic lights for faster ambulance journeys and so on

Sandra says health is one of the most exciting areas for AI, climate change too. But also it could just help us to enjoy our jobs more - taking away small annoying jobs. Eg in her job, she has a tool that automates the footnotes when she writes research papers

It’s a break now - so let’s look at some of the work that BBC R&D are exhibiting from their experimentation with AI...

...first up - the Cognitus project uses AI to augment user generated video that’s provided to the BBC by the public to be nearer broadcast quality. It might not come across clearly in my photo, but the picture on the right is much better quality than the original on the left...

...this technology is already deployed in the BBC and has been used in their coverage of the Edinburgh Festival, and the Springwatch/Winterwatxh series.
They say audiences want to be able to contribute now, and this enables their footage to be used by improving picture quality

This one is really fascinating. They’re experimenting with having AI able to film and direct. In this case trying a panel show. There are four cameras, one of which is shown here, but from each camera the AI manages a number of ‘virtual shots’ that it can select from...

To frame shots it identifies faces, facial expressions, who is speaking and so on. But it also analyses field of gaze. It uses that to understand who is important at that moment; but also to frame close up shots to have space between face and edge of screen in direction of gaze

The AI has been trained on a range of BBC panel shows to learn how directors frame and select shots. Interestingly the team also interviewed directors, but found what they said was important in decisions wasn’t supported by evidence from studying actual footage...

...that’s an interesting insight, and one user researchers will recognise. Good user research gets to what people actually do, rather than just what they say they do.

I won’t live-tweet the next session as it’s being recorded and will be broadcast on The Media Show on Radio 4 on Wednesday! Tune in for a discussion on the impact on young people growing up in a world governed by algorithms